Landscape-Scale Long-Term Drought Prevalence Mapping for Small Municipalities Adaptation, the Czech Republic Case Study

Round 1

Reviewer 1 Report (Previous Reviewer 1)

The authors have considerably changed the earlier version and improved the quality of the article. I would like to thank all the authors for this. Nevertheless, making the following corrections would make the paper better. 

 

In the paper, you build a model for 1991-2014 and use it to forecast 2021-2041 and 2041-2061. Don't the other studies you mentioned have any comparative data between 204-2021? How can we fill the data gap? Also, we do not yet know the accuracy of the data between 2021-2041 and 2041-2061 (as there are no reference values). Therefore, why don't we predict 2014-2021?

 

How can we trust the model without being able to determine its accuracy? For example, if we separate some of the values in this dataset as test data, can we better question the accuracy of the model?

Author Response

The authors have considerably changed the earlier version and improved the quality of the article. I would like to thank all the authors for this. Nevertheless, making the following corrections would make the paper better. 

 In the paper, you build a model for 1991-2014 and use it to forecast 2021-2041 and 2041-2061. Don't the other studies you mentioned have any comparative data between 204-2021? How can we fill the data gap? Also, we do not yet know the accuracy of the data between 2021-2041 and 2041-2061 (as there are no reference values). Therefore, why don't we predict 2014-2021?

The period 1991 – 2014 was used as a baseline, i.e. to represent normal climatic characteristics for the study areas. Usually 30 year periods are used for this purposes (climatic normal), but data for years 1991 – 2020 are not available. We used the closest available data, i. e. 1991 – 2014. We consider this 25 year period is representative enough and serve better for the purpose of this study than currently valid climatic normal of 1980 – 2010. This comment was also added in the manuscript.  

How can we trust the model without being able to determine its accuracy? For example, if we separate some of the values in this dataset as test data, can we better question the accuracy of the model?

The validation of the model was extensively transformed, some more diagnostics about model preformation were added to make the process more clear.

 

All edits are highlighted in the manuscript.

Thank you for your time when reviewing our manuscript.

Best regards

Authors

Reviewer 2 Report (Previous Reviewer 2)

All questions are addressed correctly. Thank you for replying to the comments and suggestions in order to improve the manuscript.

Author Response

Dear Reviewer, 

Thank you for your time when reviewing our manuscript.

Best regards

Authors

Reviewer 3 Report (Previous Reviewer 3)

The study is very necessary and well done as a modelling of landscape changes due to climate change. For this reason, I recommend publishing.

I recommend minor formal proofreading: hyphen without spaces: e.g. "1961–2014", " "spatial planning maps" (abstract, line 22).

Author Response

Dear Reviewer, we carefully read the text and corrected all formal errors. 

Reviewer 4 Report (Previous Reviewer 4)

The authors collected a number of layers like climate layers, LULC layers, and elevation layers with various spatial resolution. With random forest model, the research tried to predict drought hazard and greenery development in the period of 2021-2040 and 2041- 2060. The purpose of the result is meaningful. But based on the following the reasons, I may suggest the editor to reject the paper. I hope after revision, you can submit the updated paper to other journals.

1.       You applied random forest model on your research. But the readers and I want to know whether the performance of the model is good or not. So I need to see some validations and some indices like root mean square error (RMSE) or others to confirm that this model is nice. If the training model is not good enough, I may be skeptical about your rest of the results. In figure2, there is a model diagnose but I cannot find it out.

2.       With various spatial resolutions, you try to have some predictions with fine resolutions. You should also have some validations about whether the downscaled layers can have good accuracy. If not any validation or indices to evaluate the process of the downscaling, other researchers and I will not apply your interesting approach to other places.

Author Response

Reviewer 4

The authors collected a number of layers like climate layers, LULC layers, and elevation layers with various spatial resolution. With random forest model, the research tried to predict drought hazard and greenery development in the period of 2021-2040 and 2041- 2060. The purpose of the result is meaningful. But based on the following the reasons, I may suggest the editor to reject the paper. I hope after revision, you can submit the updated paper to other journals.

1. You applied random forest model on your research. But the readers and I want to know whether the performance of the model is good or not. So I need to see some validations and some indices like root mean square error (RMSE) or others to confirm that this model is nice. If the training model is not good enough, I may be skeptical about your rest of the results. In figure2, there is a model diagnose but I cannot find it out.

 

The validation of the model was extensively transformed, some more diagnostics about model preformation were added to make the process more clear.

 

2. With various spatial resolutions, you try to have some predictions with fine resolutions. You should also have some validations about whether the downscaled layers can have good accuracy. If not any validation or indices to evaluate the process of the downscaling, other researchers and I will not apply your interesting approach to other places.

            A part commenting on this has been added in the discussion. The RF method is based on statistical evaluation of entities’ attributes in relation to the reference layer. For this reason, the accuracy of the downscaling corresponds with the model performance (based on the rules the mode learnt during training), only entities (points) are collected from a finer fishnet and have attributes derived from layers with higher spatial detail when possible (LULC, topography).

 

All edits are highlighted in the manuscript.

Thank you for your time when reviewing our manuscript.

Best regards

Authors

 

Round 2

Reviewer 1 Report (Previous Reviewer 1)

The authors have made reasonable clarifications and significant improvements, although not all of the changes I suggested in the previous revision.

Reviewer 4 Report (Previous Reviewer 4)

Thank you for your revision and your response to my comments. Your updated version improved a lot. I will accept this form and I hope in your future study, you can make some big achievements on model downscaling and the combination of global model and regional model.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

The article is remarkable for its subject and methodology. However, I have identified some significant problems with this article and I suggest you make the following corrections. 

Line 29: Citations are not shown this way. For two citations, they should be arranged as [1,2].  I suggest you review the full article.

 

Line 33: Similarly, citations here should be cited as [6-10]. I suggest you review the full article.

 

Line 74-79: 13 different factors are mentioned here. Reference can be made to studies where these factors are discussed in detail. For example, NDVI, LST, TWI are factors that we frequently come across in the literature.

 

Line 95-96: You said that unlike many studies, the model used in your study is for a smaller area. I don't understand whether this is an advantage or a disadvantage and what is the harm of working in large areas. If you have mentioned this in the article and I have missed it, please enlighten me.

 

Line 136-201: You do not have to refer to Figure 1 everywhere. If you want to draw attention to different areas in Figure 1, define the Figure in parts (like A, B, C, D) and refer to them. But I think it is not necessary. This is already your workflow and you are explaining it.

 

Line 268: Why 1991-2014? No data until 2021? Why is there a data gap? If there is no valid reason for this, I suggest you extend your study to 2021.

 

Line 358: You should separate the Discussion and conclusion section of the paper. In the Discussion you should comment on the model, the causes and consequences of the drought, the reliability of your data and your accuracy analysis. I don't see a conclusion. This is only a small part of the discussion.

 

Line 428: This link is not working. (Page not found error)

 

Reviewer 2 Report

This Manuscript presents interesting and well-elaborated research about landscape scale drought modeling. Some minor changes are suggested to improve the Manuscript further.
 
- Please check the Instructions for Authors regarding reference numbers. For example, instead of [3],[4], [5], it should be [3-5], etc.
 
- Please check if the term “Drought risk” is adequate (lines 189, 196, 227,259, 263,…). Drought risk is usually referred to as the combination of drought vulnerability and drought hazard, and it seems that it is not in this case.

- Please check the sentence in line 216.

Reviewer 3 Report

The study is very necessary and well done as a modelling of landscape changes due to climate change. For this reason, I recommend publishing. I see a certain research deficit in the area of links to municipalities and their administration.

In the introduction or discussion, I would recommend answering the questions:

1) What is the role of (small) municipalities in planning landscape adaptation? What are their attitudes towards adaptation? What other barriers are there (e.g. economic, land ownership)?

2) Are not tools of landscape adaptation such as landscape plans ("krajinné plány"), land adjustments or revitalization of the water regime discussed?

3) How to activate the municipality administration or local citizens to adapt the landscape? What is the relationship between the adaptation of built-up areas and the rural landscape of municipalities? (See e.g. https://doi.org/10.1111/jfr3.12474, https://doi.org/10.3390/w13152098, https://doi.org/10.1080/08941920701818258) 

4) How the model areas of municipalities were chosen? According to Figure 2, these are often not extreme drought risk areas. (Figure 5: LULC of municipalities belongs more to the description of the study area -2.1)

5) What role does arable crop change or farming mode play in modelling landscape changes due to drought risk?

6) I recommend revising the formal elements: Table 2: add absolute values, Figure 6: color water bodies x scale in blue, I recommend mentioning "CzechGlobe", years e.g. "1961–2014", “spatial planning maps“ (in abstract).

 

 

Reviewer 4 Report

This paper presents a layer overlapping and random forest approach to map the drought risk in future. Specifically, the authors propose a large number of layers from many data sources. The method is easy to follow. However, based on the following reasons, I might reject the paper.

1.     The modeling is lacks of validation. When we try to run a model, we might have a training set and a validation set which provides an unbiased evaluation of a model fit on the training dataset. The authors pose a residual map to evaluate the effectiveness of model which could not avoid the overfitting of the model. Without validation, I might be doubtful about the rest of the results.

2.     The Global Circulation Models (GCMs) might not be a good choice here. After you downscaled the data created by GCM, the resolution is 500 meter. Based on layer stacking multiple layers with various spatial resolutions, the spatial resolution of your result is 20 meter (local scale) which might not be supported by the GCM. Global climate models do not have sufficient spatial resolution to represent the regional atmospheric and land surface processes (Salathe et al., 2010) let alone a local scale.

Salathé, E. P., Leung, L. R., Qian, Y., & Zhang, Y. (2010). Regional climate model projections for the State of Washington. Climatic Change, 102, 51-75.